Databento: live and historical market data for trading applications

Databento supplies market data to financial and technology firms, including broker-dealers, hedge funds, proprietary trading businesses and fintech developers. For a broker, its main uses are research, pricing inputs, charting and market-data applications. The product described here is a data service: receiving a feed does not place an order, provide a dealing counterparty or create a brokerage account.[2]

The distinction between a dataset and a schema is practical. The dataset identifies the publisher or feed; the schema determines which representation the application requests, such as trades, quotations or order-book events. Historical availability and product subscriptions must therefore be checked at that combination, rather than inferred from a general claim to cover an exchange.[4]

Nanosecond timestamp resolution describes the field’s precision, rather than guaranteeing nanosecond accuracy. Databento documents hardware receive timestamps and clock synchronisation using Precision Time Protocol, while publisher-generated timestamps retain the publisher’s own limitations. Backfilled periods can use different capture infrastructure. Applications comparing venues or measuring latency need to distinguish event time, receipt time and the relevant dataset history.[20]

Slow consumers can lose records under the live API’s default skip policy for specified stateless schemas, including MBP-1, MBP-10 and interval BBO feeds. A warning policy instead preserves the stream while the client falls behind. The policy is selected when creating the client. Separately, an unresolved symbol error no longer automatically closes the session: an open connection with no successfully resolved symbols can deliver no data. Connection health and usable feed coverage therefore require separate monitoring.[22]

Retained history limits recovery from prolonged lag. If records age out, the warning policy ends the session with REPLAY_DATA_AGED_OUT; the skip policy resumes from the earliest retained records. Preserving an initially delayed stream is therefore different from guaranteeing unlimited replay.[21]

Databento publishes dataset-specific plans. The following prices are the CME selection checked on 20 October 2026, in US dollars; they are not a universal tariff for every exchange. Exchange fees are additional where stated, and the larger subscriptions have an annual commitment despite being expressed as monthly amounts.[9]

These CME prices took effect on 22 June 2026. Existing Standard customers could retain US$179 monthly for twelve months if the subscription remained continuously active; cancelling and restarting removes that protection. The announcement says existing Plus and Unlimited customers retain their earlier prices. A legacy invoice can therefore legitimately differ from the current new-customer rate.[10]

For live data, personal use means an eligible non-professional individual’s own use. That qualification should not be extended to a brokerage’s production application or customer distribution.[16]

Historical usage is measured from the original binary data size, so converting delivery to CSV does not create a format-overhead surcharge. Repeated streaming requests are charged again. Downloading the same batch repeatedly within its thirty-day availability period does not incur another data charge.[11]

The PCAP page illustrates a CME monthly bundle at US$1,400 before tax after its stated bundle discount, and separately states a US$3,000 minimum order. This is a raw-capture example with an order minimum, rather than an all-in normalized-data subscription. The chosen dates, feeds, destination and total order determine the purchase.[5]

Billing runs by calendar month, including requests initiated through 23:59:59 UTC on the final day. The invoice is issued on the first day of the next month, with payment due five calendar days after month-end. The optional historical-usage cap stops further historical requests once the limit is exceeded; it does not cap live subscriptions. Failed payment blocks new requests and downloads of existing data. Automatic retries run for up to five business days after the payment date, after which support intervention may be needed.[12]

Exchange fees are collected in advance monthly and passed through without a vendor markup, but the documentation adds a 2.9% Stripe processing charge. The venue-fee cancellation provision should not be confused with cancellation of an annual Databento subscription.[16]

The portal catalog exposes dataset history, capture details, licence requirements, samples and API examples. Its calculator estimates usage from the previous thirty days; the cost API can calculate a specific historical request. That distinction lets a team test a representative request and budget an intended history window before scheduling a recurring research or ingestion process.[15]

Live onboarding combines a subscription with a subscriber questionnaire and the relevant exchange permissions. For commercial uses requiring a direct venue agreement, Databento submits the declaration and the venue handles its approval and paperwork. Data access follows the approved effective date. An existing licence can be supplied for verification; purchasing an API plan alone does not establish every intended use right.[16]

Databento describes self-hosted servers for its core service, with AWS used for services including public DNS, content delivery and independent monitoring. Its architecture documentation sets a twelve-month minimum term for dedicated connectivity and distinguishes its own charges from estimated third-party charges. A private circuit or cloud interconnect changes delivery, but does not by itself establish an end-to-end availability or recovery commitment.[18]

The live documentation lists planned restarts: CME on Saturdays at 02:15 Central Time, ICE on Sundays at 09:45 UTC and other datasets on Sundays at 10:30 UTC, with a recommended fifteen-minute wait. It also allows urgent midweek restarts without notice. Weekend live test data can appear without entering historical storage.[21]

The public status service separates components such as the portal, historical service, live datasets and file transfer. Component history helps identify the affected service and distinguish a venue-specific interruption from a wider outage. A current status indication is an operational observation, not a contractual uptime guarantee or evidence of the client’s own connection quality.[24]

The public User Agreement promises reasonable support for an account in good standing and commercially reasonable efforts to correct reproducible errors. It permits temporary restriction for security or emergency maintenance. Its general data and service disclaimer does not promise uninterrupted operation or complete accuracy. No universal numerical response-time or uptime commitment is established by these provisions.[27]

The agreement identifies Databento, Inc. as the operator and grants a restricted, non-transferable internal-use right subject to third-party data terms. Customers must remain qualified to receive exchange information under applicable rules. This is the permission framework for using the service; it should not be treated as a financial-services authorisation for the customer’s brokerage or investment activity.[27]

Databento dates its founding to February 2019 and the launch of its public historical-data platform to April 2023. Its company page identifies Christina Qi as co-founder and chief executive. This history helps distinguish the company’s formation from the later availability of a particular data service.[1]

On 9 July 2026, Databento announced a US$97 million Series B led by New Enterprise Associates, with participation from DRW, Redpoint Ventures and Tribe Capital. The same release described planned expansion to twenty data centres over the following six months. That expansion was a forward plan, and is not presented here as a completed October footprint.[2]

Operator

Databento, Inc.

Service scope

Live and historical normalized market data, raw PCAP, corporate actions, security master and calculated indices.

Customer types

Broker-dealers; Hedge funds; Proprietary trading firms; Fintech developers; Commodity trading firms

Delivery

API; Managed service

Pricing publication

Public and quoted pricing

Platforms

Live API; Historical API; Databento Portal

Integrations

Raw TCP live feed; HTTP historical API; Python, C++ and Rust clients; Cloud interconnects and cross-connects

Pricing scope

Public dataset subscriptions and historical usage rates; raw-data orders and dedicated network delivery have additional scope and charges.

Implementation

Select datasets and schemas, establish exchange permissions, test ingestion and plan replay, state rebuilding and download retention.

Support

Official email, chat and community routes, with a separate public component-status service. Customer-specific commitments require their applicable terms.

Permissions and regulatory scope

Market-data service with exchange and third-party licensing restrictions; receipt of data does not authorise the customer’s financial activities.

Provider metadata checked 20 Oct 2026Source 1Source 2Source 3Source 4Source 5Source 6Source 7Source 8Source 9Source 15Source 16Source 23Source 27

A change to Nasdaq TotalView-ITCH historical data is scheduled for 31 October 2026, after this profile’s check date. The 30 September notice describes synthetic instrument identifiers for older history before 28 November 2023, retention of the original Stock Locate in raw_instrument_id, restoration of missing final after-hours data through 01:00 UTC and midnight-UTC snapshots for selected schemas. A preview was available from 30 September at the existing data cost. Live data is unaffected. A team comparing historical extracts across the cutover needs to account for these identifier and completeness changes; the announced cutover is not yet a completed event as of 20 October.[26]

Continuous-futures symbology maps a rolling series to the underlying contract on each date while preserving that contract’s original prices. It is not a back-adjusted price series. Roll gaps therefore remain in the returned history, which matters when calculating returns, generating charts or testing a strategy that assumes a smooth adjusted series.[25]

Core Indices adds a different kind of input to the catalog: calculated benchmarks rather than executable quotations. The launch page gives history from 23 November 2025, subscriptions starting at US$99 monthly and historical usage starting at US$10 per gigabyte. Those starting prices belong to that product and should not be substituted for CME or another dataset’s tariff.[8]

Offboarding has two different states. Deactivating a member blocks login and removes existing integrations, but the member can be restored; deletion is permanent. Deleting the whole team requires the owner to cancel licences and settle balances first, then immediately and permanently removes portal and API access for every member. Exporting required data and completing application handover must precede that account action.[14]